2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
18 #include <linux/log2.h>
22 #include "xfs_shared.h"
23 #include "xfs_format.h"
24 #include "xfs_log_format.h"
25 #include "xfs_trans_resv.h"
29 #include "xfs_mount.h"
30 #include "xfs_da_format.h"
31 #include "xfs_da_btree.h"
33 #include "xfs_bmap_btree.h"
34 #include "xfs_alloc_btree.h"
35 #include "xfs_ialloc_btree.h"
36 #include "xfs_attr_sf.h"
38 #include "xfs_dinode.h"
39 #include "xfs_inode.h"
40 #include "xfs_trans_space.h"
41 #include "xfs_trans.h"
42 #include "xfs_buf_item.h"
43 #include "xfs_inode_item.h"
44 #include "xfs_btree.h"
45 #include "xfs_alloc.h"
46 #include "xfs_ialloc.h"
48 #include "xfs_bmap_util.h"
49 #include "xfs_error.h"
50 #include "xfs_quota.h"
51 #include "xfs_filestream.h"
52 #include "xfs_cksum.h"
53 #include "xfs_trace.h"
54 #include "xfs_icache.h"
55 #include "xfs_symlink.h"
56 #include "xfs_trans_priv.h"
59 kmem_zone_t *xfs_inode_zone;
62 * Used in xfs_itruncate_extents(). This is the maximum number of extents
63 * freed from a file in a single transaction.
65 #define XFS_ITRUNC_MAX_EXTENTS 2
67 STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *);
70 * helper function to extract extent size hint from inode
76 if ((ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE) && ip->i_d.di_extsize)
77 return ip->i_d.di_extsize;
78 if (XFS_IS_REALTIME_INODE(ip))
79 return ip->i_mount->m_sb.sb_rextsize;
84 * This is a wrapper routine around the xfs_ilock() routine used to centralize
85 * some grungy code. It is used in places that wish to lock the inode solely
86 * for reading the extents. The reason these places can't just call
87 * xfs_ilock(SHARED) is that the inode lock also guards to bringing in of the
88 * extents from disk for a file in b-tree format. If the inode is in b-tree
89 * format, then we need to lock the inode exclusively until the extents are read
90 * in. Locking it exclusively all the time would limit our parallelism
91 * unnecessarily, though. What we do instead is check to see if the extents
92 * have been read in yet, and only lock the inode exclusively if they have not.
94 * The function returns a value which should be given to the corresponding
95 * xfs_iunlock_map_shared(). This value is the mode in which the lock was
104 if ((ip->i_d.di_format == XFS_DINODE_FMT_BTREE) &&
105 ((ip->i_df.if_flags & XFS_IFEXTENTS) == 0)) {
106 lock_mode = XFS_ILOCK_EXCL;
108 lock_mode = XFS_ILOCK_SHARED;
111 xfs_ilock(ip, lock_mode);
117 * This is simply the unlock routine to go with xfs_ilock_map_shared().
118 * All it does is call xfs_iunlock() with the given lock_mode.
121 xfs_iunlock_map_shared(
123 unsigned int lock_mode)
125 xfs_iunlock(ip, lock_mode);
129 * The xfs inode contains 2 locks: a multi-reader lock called the
130 * i_iolock and a multi-reader lock called the i_lock. This routine
131 * allows either or both of the locks to be obtained.
133 * The 2 locks should always be ordered so that the IO lock is
134 * obtained first in order to prevent deadlock.
136 * ip -- the inode being locked
137 * lock_flags -- this parameter indicates the inode's locks
138 * to be locked. It can be:
143 * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED,
144 * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL,
145 * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED,
146 * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL
153 trace_xfs_ilock(ip, lock_flags, _RET_IP_);
156 * You can't set both SHARED and EXCL for the same lock,
157 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
158 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
160 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
161 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
162 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
163 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
164 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
166 if (lock_flags & XFS_IOLOCK_EXCL)
167 mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
168 else if (lock_flags & XFS_IOLOCK_SHARED)
169 mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags));
171 if (lock_flags & XFS_ILOCK_EXCL)
172 mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
173 else if (lock_flags & XFS_ILOCK_SHARED)
174 mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
178 * This is just like xfs_ilock(), except that the caller
179 * is guaranteed not to sleep. It returns 1 if it gets
180 * the requested locks and 0 otherwise. If the IO lock is
181 * obtained but the inode lock cannot be, then the IO lock
182 * is dropped before returning.
184 * ip -- the inode being locked
185 * lock_flags -- this parameter indicates the inode's locks to be
186 * to be locked. See the comment for xfs_ilock() for a list
194 trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_);
197 * You can't set both SHARED and EXCL for the same lock,
198 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
199 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
201 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
202 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
203 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
204 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
205 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
207 if (lock_flags & XFS_IOLOCK_EXCL) {
208 if (!mrtryupdate(&ip->i_iolock))
210 } else if (lock_flags & XFS_IOLOCK_SHARED) {
211 if (!mrtryaccess(&ip->i_iolock))
214 if (lock_flags & XFS_ILOCK_EXCL) {
215 if (!mrtryupdate(&ip->i_lock))
216 goto out_undo_iolock;
217 } else if (lock_flags & XFS_ILOCK_SHARED) {
218 if (!mrtryaccess(&ip->i_lock))
219 goto out_undo_iolock;
224 if (lock_flags & XFS_IOLOCK_EXCL)
225 mrunlock_excl(&ip->i_iolock);
226 else if (lock_flags & XFS_IOLOCK_SHARED)
227 mrunlock_shared(&ip->i_iolock);
233 * xfs_iunlock() is used to drop the inode locks acquired with
234 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
235 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
236 * that we know which locks to drop.
238 * ip -- the inode being unlocked
239 * lock_flags -- this parameter indicates the inode's locks to be
240 * to be unlocked. See the comment for xfs_ilock() for a list
241 * of valid values for this parameter.
250 * You can't set both SHARED and EXCL for the same lock,
251 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
252 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
254 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
255 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
256 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
257 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
258 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0);
259 ASSERT(lock_flags != 0);
261 if (lock_flags & XFS_IOLOCK_EXCL)
262 mrunlock_excl(&ip->i_iolock);
263 else if (lock_flags & XFS_IOLOCK_SHARED)
264 mrunlock_shared(&ip->i_iolock);
266 if (lock_flags & XFS_ILOCK_EXCL)
267 mrunlock_excl(&ip->i_lock);
268 else if (lock_flags & XFS_ILOCK_SHARED)
269 mrunlock_shared(&ip->i_lock);
271 trace_xfs_iunlock(ip, lock_flags, _RET_IP_);
275 * give up write locks. the i/o lock cannot be held nested
276 * if it is being demoted.
283 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL));
284 ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
286 if (lock_flags & XFS_ILOCK_EXCL)
287 mrdemote(&ip->i_lock);
288 if (lock_flags & XFS_IOLOCK_EXCL)
289 mrdemote(&ip->i_iolock);
291 trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_);
294 #if defined(DEBUG) || defined(XFS_WARN)
300 if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) {
301 if (!(lock_flags & XFS_ILOCK_SHARED))
302 return !!ip->i_lock.mr_writer;
303 return rwsem_is_locked(&ip->i_lock.mr_lock);
306 if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) {
307 if (!(lock_flags & XFS_IOLOCK_SHARED))
308 return !!ip->i_iolock.mr_writer;
309 return rwsem_is_locked(&ip->i_iolock.mr_lock);
319 int xfs_small_retries;
320 int xfs_middle_retries;
321 int xfs_lots_retries;
326 * Bump the subclass so xfs_lock_inodes() acquires each lock with
330 xfs_lock_inumorder(int lock_mode, int subclass)
332 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
333 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_IOLOCK_SHIFT;
334 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))
335 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_ILOCK_SHIFT;
341 * The following routine will lock n inodes in exclusive mode.
342 * We assume the caller calls us with the inodes in i_ino order.
344 * We need to detect deadlock where an inode that we lock
345 * is in the AIL and we start waiting for another inode that is locked
346 * by a thread in a long running transaction (such as truncate). This can
347 * result in deadlock since the long running trans might need to wait
348 * for the inode we just locked in order to push the tail and free space
357 int attempts = 0, i, j, try_lock;
360 ASSERT(ips && (inodes >= 2)); /* we need at least two */
366 for (; i < inodes; i++) {
369 if (i && (ips[i] == ips[i-1])) /* Already locked */
373 * If try_lock is not set yet, make sure all locked inodes
374 * are not in the AIL.
375 * If any are, set try_lock to be used later.
379 for (j = (i - 1); j >= 0 && !try_lock; j--) {
380 lp = (xfs_log_item_t *)ips[j]->i_itemp;
381 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
388 * If any of the previous locks we have locked is in the AIL,
389 * we must TRY to get the second and subsequent locks. If
390 * we can't get any, we must release all we have
395 /* try_lock must be 0 if i is 0. */
397 * try_lock means we have an inode locked
398 * that is in the AIL.
401 if (!xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) {
405 * Unlock all previous guys and try again.
406 * xfs_iunlock will try to push the tail
407 * if the inode is in the AIL.
410 for(j = i - 1; j >= 0; j--) {
413 * Check to see if we've already
415 * Not the first one going back,
416 * and the inode ptr is the same.
418 if ((j != (i - 1)) && ips[j] ==
422 xfs_iunlock(ips[j], lock_mode);
425 if ((attempts % 5) == 0) {
426 delay(1); /* Don't just spin the CPU */
436 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
442 if (attempts < 5) xfs_small_retries++;
443 else if (attempts < 100) xfs_middle_retries++;
444 else xfs_lots_retries++;
452 * xfs_lock_two_inodes() can only be used to lock one type of lock
453 * at a time - the iolock or the ilock, but not both at once. If
454 * we lock both at once, lockdep will report false positives saying
455 * we have violated locking orders.
467 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))
468 ASSERT((lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) == 0);
469 ASSERT(ip0->i_ino != ip1->i_ino);
471 if (ip0->i_ino > ip1->i_ino) {
478 xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0));
481 * If the first lock we have locked is in the AIL, we must TRY to get
482 * the second lock. If we can't get it, we must release the first one
485 lp = (xfs_log_item_t *)ip0->i_itemp;
486 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
487 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) {
488 xfs_iunlock(ip0, lock_mode);
489 if ((++attempts % 5) == 0)
490 delay(1); /* Don't just spin the CPU */
494 xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1));
501 struct xfs_inode *ip)
503 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT);
504 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT);
507 prepare_to_wait_exclusive(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
508 if (xfs_isiflocked(ip))
510 } while (!xfs_iflock_nowait(ip));
512 finish_wait(wq, &wait.wait);
521 if (di_flags & XFS_DIFLAG_ANY) {
522 if (di_flags & XFS_DIFLAG_REALTIME)
523 flags |= XFS_XFLAG_REALTIME;
524 if (di_flags & XFS_DIFLAG_PREALLOC)
525 flags |= XFS_XFLAG_PREALLOC;
526 if (di_flags & XFS_DIFLAG_IMMUTABLE)
527 flags |= XFS_XFLAG_IMMUTABLE;
528 if (di_flags & XFS_DIFLAG_APPEND)
529 flags |= XFS_XFLAG_APPEND;
530 if (di_flags & XFS_DIFLAG_SYNC)
531 flags |= XFS_XFLAG_SYNC;
532 if (di_flags & XFS_DIFLAG_NOATIME)
533 flags |= XFS_XFLAG_NOATIME;
534 if (di_flags & XFS_DIFLAG_NODUMP)
535 flags |= XFS_XFLAG_NODUMP;
536 if (di_flags & XFS_DIFLAG_RTINHERIT)
537 flags |= XFS_XFLAG_RTINHERIT;
538 if (di_flags & XFS_DIFLAG_PROJINHERIT)
539 flags |= XFS_XFLAG_PROJINHERIT;
540 if (di_flags & XFS_DIFLAG_NOSYMLINKS)
541 flags |= XFS_XFLAG_NOSYMLINKS;
542 if (di_flags & XFS_DIFLAG_EXTSIZE)
543 flags |= XFS_XFLAG_EXTSIZE;
544 if (di_flags & XFS_DIFLAG_EXTSZINHERIT)
545 flags |= XFS_XFLAG_EXTSZINHERIT;
546 if (di_flags & XFS_DIFLAG_NODEFRAG)
547 flags |= XFS_XFLAG_NODEFRAG;
548 if (di_flags & XFS_DIFLAG_FILESTREAM)
549 flags |= XFS_XFLAG_FILESTREAM;
559 xfs_icdinode_t *dic = &ip->i_d;
561 return _xfs_dic2xflags(dic->di_flags) |
562 (XFS_IFORK_Q(ip) ? XFS_XFLAG_HASATTR : 0);
569 return _xfs_dic2xflags(be16_to_cpu(dip->di_flags)) |
570 (XFS_DFORK_Q(dip) ? XFS_XFLAG_HASATTR : 0);
574 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
575 * is allowed, otherwise it has to be an exact match. If a CI match is found,
576 * ci_name->name will point to a the actual name (caller must free) or
577 * will be set to NULL if an exact match is found.
582 struct xfs_name *name,
584 struct xfs_name *ci_name)
590 trace_xfs_lookup(dp, name);
592 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
593 return XFS_ERROR(EIO);
595 lock_mode = xfs_ilock_map_shared(dp);
596 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
597 xfs_iunlock_map_shared(dp, lock_mode);
602 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
610 kmem_free(ci_name->name);
617 * Allocate an inode on disk and return a copy of its in-core version.
618 * The in-core inode is locked exclusively. Set mode, nlink, and rdev
619 * appropriately within the inode. The uid and gid for the inode are
620 * set according to the contents of the given cred structure.
622 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc()
623 * has a free inode available, call xfs_iget() to obtain the in-core
624 * version of the allocated inode. Finally, fill in the inode and
625 * log its initial contents. In this case, ialloc_context would be
628 * If xfs_dialloc() does not have an available inode, it will replenish
629 * its supply by doing an allocation. Since we can only do one
630 * allocation within a transaction without deadlocks, we must commit
631 * the current transaction before returning the inode itself.
632 * In this case, therefore, we will set ialloc_context and return.
633 * The caller should then commit the current transaction, start a new
634 * transaction, and call xfs_ialloc() again to actually get the inode.
636 * To ensure that some other process does not grab the inode that
637 * was allocated during the first call to xfs_ialloc(), this routine
638 * also returns the [locked] bp pointing to the head of the freelist
639 * as ialloc_context. The caller should hold this buffer across
640 * the commit and pass it back into this routine on the second call.
642 * If we are allocating quota inodes, we do not have a parent inode
643 * to attach to or associate with (i.e. pip == NULL) because they
644 * are not linked into the directory structure - they are attached
645 * directly to the superblock - and so have no parent.
656 xfs_buf_t **ialloc_context,
659 struct xfs_mount *mp = tp->t_mountp;
668 * Call the space management code to pick
669 * the on-disk inode to be allocated.
671 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc,
672 ialloc_context, &ino);
675 if (*ialloc_context || ino == NULLFSINO) {
679 ASSERT(*ialloc_context == NULL);
682 * Get the in-core inode with the lock held exclusively.
683 * This is because we're setting fields here we need
684 * to prevent others from looking at until we're done.
686 error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE,
687 XFS_ILOCK_EXCL, &ip);
692 ip->i_d.di_mode = mode;
693 ip->i_d.di_onlink = 0;
694 ip->i_d.di_nlink = nlink;
695 ASSERT(ip->i_d.di_nlink == nlink);
696 ip->i_d.di_uid = xfs_kuid_to_uid(current_fsuid());
697 ip->i_d.di_gid = xfs_kgid_to_gid(current_fsgid());
698 xfs_set_projid(ip, prid);
699 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
702 * If the superblock version is up to where we support new format
703 * inodes and this is currently an old format inode, then change
704 * the inode version number now. This way we only do the conversion
705 * here rather than here and in the flush/logging code.
707 if (xfs_sb_version_hasnlink(&mp->m_sb) &&
708 ip->i_d.di_version == 1) {
709 ip->i_d.di_version = 2;
711 * We've already zeroed the old link count, the projid field,
717 * Project ids won't be stored on disk if we are using a version 1 inode.
719 if ((prid != 0) && (ip->i_d.di_version == 1))
720 xfs_bump_ino_vers2(tp, ip);
722 if (pip && XFS_INHERIT_GID(pip)) {
723 ip->i_d.di_gid = pip->i_d.di_gid;
724 if ((pip->i_d.di_mode & S_ISGID) && S_ISDIR(mode)) {
725 ip->i_d.di_mode |= S_ISGID;
730 * If the group ID of the new file does not match the effective group
731 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared
732 * (and only if the irix_sgid_inherit compatibility variable is set).
734 if ((irix_sgid_inherit) &&
735 (ip->i_d.di_mode & S_ISGID) &&
736 (!in_group_p(xfs_gid_to_kgid(ip->i_d.di_gid)))) {
737 ip->i_d.di_mode &= ~S_ISGID;
741 ip->i_d.di_nextents = 0;
742 ASSERT(ip->i_d.di_nblocks == 0);
745 ip->i_d.di_mtime.t_sec = (__int32_t)tv.tv_sec;
746 ip->i_d.di_mtime.t_nsec = (__int32_t)tv.tv_nsec;
747 ip->i_d.di_atime = ip->i_d.di_mtime;
748 ip->i_d.di_ctime = ip->i_d.di_mtime;
751 * di_gen will have been taken care of in xfs_iread.
753 ip->i_d.di_extsize = 0;
754 ip->i_d.di_dmevmask = 0;
755 ip->i_d.di_dmstate = 0;
756 ip->i_d.di_flags = 0;
758 if (ip->i_d.di_version == 3) {
759 ASSERT(ip->i_d.di_ino == ino);
760 ASSERT(uuid_equal(&ip->i_d.di_uuid, &mp->m_sb.sb_uuid));
762 ip->i_d.di_changecount = 1;
764 ip->i_d.di_flags2 = 0;
765 memset(&(ip->i_d.di_pad2[0]), 0, sizeof(ip->i_d.di_pad2));
766 ip->i_d.di_crtime = ip->i_d.di_mtime;
770 flags = XFS_ILOG_CORE;
771 switch (mode & S_IFMT) {
776 ip->i_d.di_format = XFS_DINODE_FMT_DEV;
777 ip->i_df.if_u2.if_rdev = rdev;
778 ip->i_df.if_flags = 0;
779 flags |= XFS_ILOG_DEV;
783 * we can't set up filestreams until after the VFS inode
784 * is set up properly.
786 if (pip && xfs_inode_is_filestream(pip))
790 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) {
794 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
795 di_flags |= XFS_DIFLAG_RTINHERIT;
796 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
797 di_flags |= XFS_DIFLAG_EXTSZINHERIT;
798 ip->i_d.di_extsize = pip->i_d.di_extsize;
800 } else if (S_ISREG(mode)) {
801 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT)
802 di_flags |= XFS_DIFLAG_REALTIME;
803 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) {
804 di_flags |= XFS_DIFLAG_EXTSIZE;
805 ip->i_d.di_extsize = pip->i_d.di_extsize;
808 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) &&
810 di_flags |= XFS_DIFLAG_NOATIME;
811 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) &&
813 di_flags |= XFS_DIFLAG_NODUMP;
814 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) &&
816 di_flags |= XFS_DIFLAG_SYNC;
817 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) &&
818 xfs_inherit_nosymlinks)
819 di_flags |= XFS_DIFLAG_NOSYMLINKS;
820 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
821 di_flags |= XFS_DIFLAG_PROJINHERIT;
822 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) &&
823 xfs_inherit_nodefrag)
824 di_flags |= XFS_DIFLAG_NODEFRAG;
825 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM)
826 di_flags |= XFS_DIFLAG_FILESTREAM;
827 ip->i_d.di_flags |= di_flags;
831 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
832 ip->i_df.if_flags = XFS_IFEXTENTS;
833 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0;
834 ip->i_df.if_u1.if_extents = NULL;
840 * Attribute fork settings for new inode.
842 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
843 ip->i_d.di_anextents = 0;
846 * Log the new values stuffed into the inode.
848 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
849 xfs_trans_log_inode(tp, ip, flags);
851 /* now that we have an i_mode we can setup inode ops and unlock */
854 /* now we have set up the vfs inode we can associate the filestream */
856 error = xfs_filestream_associate(pip, ip);
860 xfs_iflags_set(ip, XFS_IFILESTREAM);
868 * Allocates a new inode from disk and return a pointer to the
869 * incore copy. This routine will internally commit the current
870 * transaction and allocate a new one if the Space Manager needed
871 * to do an allocation to replenish the inode free-list.
873 * This routine is designed to be called from xfs_create and
879 xfs_trans_t **tpp, /* input: current transaction;
880 output: may be a new transaction. */
881 xfs_inode_t *dp, /* directory within whose allocate
886 prid_t prid, /* project id */
887 int okalloc, /* ok to allocate new space */
888 xfs_inode_t **ipp, /* pointer to inode; it will be
896 xfs_buf_t *ialloc_context = NULL;
902 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
905 * xfs_ialloc will return a pointer to an incore inode if
906 * the Space Manager has an available inode on the free
907 * list. Otherwise, it will do an allocation and replenish
908 * the freelist. Since we can only do one allocation per
909 * transaction without deadlocks, we will need to commit the
910 * current transaction and start a new one. We will then
911 * need to call xfs_ialloc again to get the inode.
913 * If xfs_ialloc did an allocation to replenish the freelist,
914 * it returns the bp containing the head of the freelist as
915 * ialloc_context. We will hold a lock on it across the
916 * transaction commit so that no other process can steal
917 * the inode(s) that we've just allocated.
919 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, okalloc,
920 &ialloc_context, &ip);
923 * Return an error if we were unable to allocate a new inode.
924 * This should only happen if we run out of space on disk or
925 * encounter a disk error.
931 if (!ialloc_context && !ip) {
933 return XFS_ERROR(ENOSPC);
937 * If the AGI buffer is non-NULL, then we were unable to get an
938 * inode in one operation. We need to commit the current
939 * transaction and call xfs_ialloc() again. It is guaranteed
940 * to succeed the second time.
942 if (ialloc_context) {
943 struct xfs_trans_res tres;
946 * Normally, xfs_trans_commit releases all the locks.
947 * We call bhold to hang on to the ialloc_context across
948 * the commit. Holding this buffer prevents any other
949 * processes from doing any allocations in this
952 xfs_trans_bhold(tp, ialloc_context);
954 * Save the log reservation so we can use
955 * them in the next transaction.
957 tres.tr_logres = xfs_trans_get_log_res(tp);
958 tres.tr_logcount = xfs_trans_get_log_count(tp);
961 * We want the quota changes to be associated with the next
962 * transaction, NOT this one. So, detach the dqinfo from this
963 * and attach it to the next transaction.
968 dqinfo = (void *)tp->t_dqinfo;
970 tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY;
971 tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY);
974 ntp = xfs_trans_dup(tp);
975 code = xfs_trans_commit(tp, 0);
977 if (committed != NULL) {
981 * If we get an error during the commit processing,
982 * release the buffer that is still held and return
986 xfs_buf_relse(ialloc_context);
988 tp->t_dqinfo = dqinfo;
989 xfs_trans_free_dqinfo(tp);
997 * transaction commit worked ok so we can drop the extra ticket
998 * reference that we gained in xfs_trans_dup()
1000 xfs_log_ticket_put(tp->t_ticket);
1001 tres.tr_logflags = XFS_TRANS_PERM_LOG_RES;
1002 code = xfs_trans_reserve(tp, &tres, 0, 0);
1005 * Re-attach the quota info that we detached from prev trx.
1008 tp->t_dqinfo = dqinfo;
1009 tp->t_flags |= tflags;
1013 xfs_buf_relse(ialloc_context);
1018 xfs_trans_bjoin(tp, ialloc_context);
1021 * Call ialloc again. Since we've locked out all
1022 * other allocations in this allocation group,
1023 * this call should always succeed.
1025 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid,
1026 okalloc, &ialloc_context, &ip);
1029 * If we get an error at this point, return to the caller
1030 * so that the current transaction can be aborted.
1037 ASSERT(!ialloc_context && ip);
1040 if (committed != NULL)
1051 * Decrement the link count on an inode & log the change.
1052 * If this causes the link count to go to zero, initiate the
1053 * logging activity required to truncate a file.
1062 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1064 ASSERT (ip->i_d.di_nlink > 0);
1066 drop_nlink(VFS_I(ip));
1067 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1070 if (ip->i_d.di_nlink == 0) {
1072 * We're dropping the last link to this file.
1073 * Move the on-disk inode to the AGI unlinked list.
1074 * From xfs_inactive() we will pull the inode from
1075 * the list and free it.
1077 error = xfs_iunlink(tp, ip);
1083 * This gets called when the inode's version needs to be changed from 1 to 2.
1084 * Currently this happens when the nlink field overflows the old 16-bit value
1085 * or when chproj is called to change the project for the first time.
1086 * As a side effect the superblock version will also get rev'd
1087 * to contain the NLINK bit.
1096 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1097 ASSERT(ip->i_d.di_version == 1);
1099 ip->i_d.di_version = 2;
1100 ip->i_d.di_onlink = 0;
1101 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
1103 if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
1104 spin_lock(&mp->m_sb_lock);
1105 if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
1106 xfs_sb_version_addnlink(&mp->m_sb);
1107 spin_unlock(&mp->m_sb_lock);
1108 xfs_mod_sb(tp, XFS_SB_VERSIONNUM);
1110 spin_unlock(&mp->m_sb_lock);
1113 /* Caller must log the inode */
1117 * Increment the link count on an inode & log the change.
1124 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
1126 ASSERT(ip->i_d.di_nlink > 0);
1128 inc_nlink(VFS_I(ip));
1129 if ((ip->i_d.di_version == 1) &&
1130 (ip->i_d.di_nlink > XFS_MAXLINK_1)) {
1132 * The inode has increased its number of links beyond
1133 * what can fit in an old format inode. It now needs
1134 * to be converted to a version 2 inode with a 32 bit
1135 * link count. If this is the first inode in the file
1136 * system to do this, then we need to bump the superblock
1137 * version number as well.
1139 xfs_bump_ino_vers2(tp, ip);
1142 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1149 struct xfs_name *name,
1154 int is_dir = S_ISDIR(mode);
1155 struct xfs_mount *mp = dp->i_mount;
1156 struct xfs_inode *ip = NULL;
1157 struct xfs_trans *tp = NULL;
1159 xfs_bmap_free_t free_list;
1160 xfs_fsblock_t first_block;
1161 bool unlock_dp_on_error = false;
1165 struct xfs_dquot *udqp = NULL;
1166 struct xfs_dquot *gdqp = NULL;
1167 struct xfs_dquot *pdqp = NULL;
1168 struct xfs_trans_res tres;
1171 trace_xfs_create(dp, name);
1173 if (XFS_FORCED_SHUTDOWN(mp))
1174 return XFS_ERROR(EIO);
1176 if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT)
1177 prid = xfs_get_projid(dp);
1179 prid = XFS_PROJID_DEFAULT;
1182 * Make sure that we have allocated dquot(s) on disk.
1184 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()),
1185 xfs_kgid_to_gid(current_fsgid()), prid,
1186 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
1187 &udqp, &gdqp, &pdqp);
1193 resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
1194 tres.tr_logres = M_RES(mp)->tr_mkdir.tr_logres;
1195 tres.tr_logcount = XFS_MKDIR_LOG_COUNT;
1196 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
1198 resblks = XFS_CREATE_SPACE_RES(mp, name->len);
1199 tres.tr_logres = M_RES(mp)->tr_create.tr_logres;
1200 tres.tr_logcount = XFS_CREATE_LOG_COUNT;
1201 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1204 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1207 * Initially assume that the file does not exist and
1208 * reserve the resources for that case. If that is not
1209 * the case we'll drop the one we have and get a more
1210 * appropriate transaction later.
1212 tres.tr_logflags = XFS_TRANS_PERM_LOG_RES;
1213 error = xfs_trans_reserve(tp, &tres, resblks, 0);
1214 if (error == ENOSPC) {
1215 /* flush outstanding delalloc blocks and retry */
1216 xfs_flush_inodes(mp);
1217 error = xfs_trans_reserve(tp, &tres, resblks, 0);
1219 if (error == ENOSPC) {
1220 /* No space at all so try a "no-allocation" reservation */
1222 error = xfs_trans_reserve(tp, &tres, 0, 0);
1226 goto out_trans_cancel;
1229 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
1230 unlock_dp_on_error = true;
1232 xfs_bmap_init(&free_list, &first_block);
1235 * Reserve disk quota and the inode.
1237 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp,
1238 pdqp, resblks, 1, 0);
1240 goto out_trans_cancel;
1242 error = xfs_dir_canenter(tp, dp, name, resblks);
1244 goto out_trans_cancel;
1247 * A newly created regular or special file just has one directory
1248 * entry pointing to them, but a directory also the "." entry
1249 * pointing to itself.
1251 error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev,
1252 prid, resblks > 0, &ip, &committed);
1254 if (error == ENOSPC)
1255 goto out_trans_cancel;
1256 goto out_trans_abort;
1260 * Now we join the directory inode to the transaction. We do not do it
1261 * earlier because xfs_dir_ialloc might commit the previous transaction
1262 * (and release all the locks). An error from here on will result in
1263 * the transaction cancel unlocking dp so don't do it explicitly in the
1266 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
1267 unlock_dp_on_error = false;
1269 error = xfs_dir_createname(tp, dp, name, ip->i_ino,
1270 &first_block, &free_list, resblks ?
1271 resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
1273 ASSERT(error != ENOSPC);
1274 goto out_trans_abort;
1276 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1277 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1280 error = xfs_dir_init(tp, ip, dp);
1282 goto out_bmap_cancel;
1284 error = xfs_bumplink(tp, dp);
1286 goto out_bmap_cancel;
1290 * If this is a synchronous mount, make sure that the
1291 * create transaction goes to disk before returning to
1294 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
1295 xfs_trans_set_sync(tp);
1298 * Attach the dquot(s) to the inodes and modify them incore.
1299 * These ids of the inode couldn't have changed since the new
1300 * inode has been locked ever since it was created.
1302 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
1304 error = xfs_bmap_finish(&tp, &free_list, &committed);
1306 goto out_bmap_cancel;
1308 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1310 goto out_release_inode;
1312 xfs_qm_dqrele(udqp);
1313 xfs_qm_dqrele(gdqp);
1314 xfs_qm_dqrele(pdqp);
1320 xfs_bmap_cancel(&free_list);
1322 cancel_flags |= XFS_TRANS_ABORT;
1324 xfs_trans_cancel(tp, cancel_flags);
1327 * Wait until after the current transaction is aborted to
1328 * release the inode. This prevents recursive transactions
1329 * and deadlocks from xfs_inactive.
1334 xfs_qm_dqrele(udqp);
1335 xfs_qm_dqrele(gdqp);
1336 xfs_qm_dqrele(pdqp);
1338 if (unlock_dp_on_error)
1339 xfs_iunlock(dp, XFS_ILOCK_EXCL);
1347 struct xfs_name *target_name)
1349 xfs_mount_t *mp = tdp->i_mount;
1352 xfs_bmap_free_t free_list;
1353 xfs_fsblock_t first_block;
1358 trace_xfs_link(tdp, target_name);
1360 ASSERT(!S_ISDIR(sip->i_d.di_mode));
1362 if (XFS_FORCED_SHUTDOWN(mp))
1363 return XFS_ERROR(EIO);
1365 error = xfs_qm_dqattach(sip, 0);
1369 error = xfs_qm_dqattach(tdp, 0);
1373 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
1374 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1375 resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
1376 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_link, resblks, 0);
1377 if (error == ENOSPC) {
1379 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_link, 0, 0);
1386 xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL);
1388 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
1389 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
1392 * If we are using project inheritance, we only allow hard link
1393 * creation in our tree when the project IDs are the same; else
1394 * the tree quota mechanism could be circumvented.
1396 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1397 (xfs_get_projid(tdp) != xfs_get_projid(sip)))) {
1398 error = XFS_ERROR(EXDEV);
1402 error = xfs_dir_canenter(tp, tdp, target_name, resblks);
1406 xfs_bmap_init(&free_list, &first_block);
1408 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
1409 &first_block, &free_list, resblks);
1412 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
1413 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
1415 error = xfs_bumplink(tp, sip);
1420 * If this is a synchronous mount, make sure that the
1421 * link transaction goes to disk before returning to
1424 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
1425 xfs_trans_set_sync(tp);
1428 error = xfs_bmap_finish (&tp, &free_list, &committed);
1430 xfs_bmap_cancel(&free_list);
1434 return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1437 cancel_flags |= XFS_TRANS_ABORT;
1439 xfs_trans_cancel(tp, cancel_flags);
1445 * Free up the underlying blocks past new_size. The new size must be smaller
1446 * than the current size. This routine can be used both for the attribute and
1447 * data fork, and does not modify the inode size, which is left to the caller.
1449 * The transaction passed to this routine must have made a permanent log
1450 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
1451 * given transaction and start new ones, so make sure everything involved in
1452 * the transaction is tidy before calling here. Some transaction will be
1453 * returned to the caller to be committed. The incoming transaction must
1454 * already include the inode, and both inode locks must be held exclusively.
1455 * The inode must also be "held" within the transaction. On return the inode
1456 * will be "held" within the returned transaction. This routine does NOT
1457 * require any disk space to be reserved for it within the transaction.
1459 * If we get an error, we must return with the inode locked and linked into the
1460 * current transaction. This keeps things simple for the higher level code,
1461 * because it always knows that the inode is locked and held in the transaction
1462 * that returns to it whether errors occur or not. We don't mark the inode
1463 * dirty on error so that transactions can be easily aborted if possible.
1466 xfs_itruncate_extents(
1467 struct xfs_trans **tpp,
1468 struct xfs_inode *ip,
1470 xfs_fsize_t new_size)
1472 struct xfs_mount *mp = ip->i_mount;
1473 struct xfs_trans *tp = *tpp;
1474 struct xfs_trans *ntp;
1475 xfs_bmap_free_t free_list;
1476 xfs_fsblock_t first_block;
1477 xfs_fileoff_t first_unmap_block;
1478 xfs_fileoff_t last_block;
1479 xfs_filblks_t unmap_len;
1484 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1485 ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
1486 xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1487 ASSERT(new_size <= XFS_ISIZE(ip));
1488 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1489 ASSERT(ip->i_itemp != NULL);
1490 ASSERT(ip->i_itemp->ili_lock_flags == 0);
1491 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1493 trace_xfs_itruncate_extents_start(ip, new_size);
1496 * Since it is possible for space to become allocated beyond
1497 * the end of the file (in a crash where the space is allocated
1498 * but the inode size is not yet updated), simply remove any
1499 * blocks which show up between the new EOF and the maximum
1500 * possible file size. If the first block to be removed is
1501 * beyond the maximum file size (ie it is the same as last_block),
1502 * then there is nothing to do.
1504 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
1505 last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
1506 if (first_unmap_block == last_block)
1509 ASSERT(first_unmap_block < last_block);
1510 unmap_len = last_block - first_unmap_block + 1;
1512 xfs_bmap_init(&free_list, &first_block);
1513 error = xfs_bunmapi(tp, ip,
1514 first_unmap_block, unmap_len,
1515 xfs_bmapi_aflag(whichfork),
1516 XFS_ITRUNC_MAX_EXTENTS,
1517 &first_block, &free_list,
1520 goto out_bmap_cancel;
1523 * Duplicate the transaction that has the permanent
1524 * reservation and commit the old transaction.
1526 error = xfs_bmap_finish(&tp, &free_list, &committed);
1528 xfs_trans_ijoin(tp, ip, 0);
1530 goto out_bmap_cancel;
1534 * Mark the inode dirty so it will be logged and
1535 * moved forward in the log as part of every commit.
1537 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1540 ntp = xfs_trans_dup(tp);
1541 error = xfs_trans_commit(tp, 0);
1544 xfs_trans_ijoin(tp, ip, 0);
1550 * Transaction commit worked ok so we can drop the extra ticket
1551 * reference that we gained in xfs_trans_dup()
1553 xfs_log_ticket_put(tp->t_ticket);
1554 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
1560 * Always re-log the inode so that our permanent transaction can keep
1561 * on rolling it forward in the log.
1563 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1565 trace_xfs_itruncate_extents_end(ip, new_size);
1572 * If the bunmapi call encounters an error, return to the caller where
1573 * the transaction can be properly aborted. We just need to make sure
1574 * we're not holding any resources that we were not when we came in.
1576 xfs_bmap_cancel(&free_list);
1584 xfs_mount_t *mp = ip->i_mount;
1587 if (!S_ISREG(ip->i_d.di_mode) || (ip->i_d.di_mode == 0))
1590 /* If this is a read-only mount, don't do this (would generate I/O) */
1591 if (mp->m_flags & XFS_MOUNT_RDONLY)
1594 if (!XFS_FORCED_SHUTDOWN(mp)) {
1598 * If we are using filestreams, and we have an unlinked
1599 * file that we are processing the last close on, then nothing
1600 * will be able to reopen and write to this file. Purge this
1601 * inode from the filestreams cache so that it doesn't delay
1602 * teardown of the inode.
1604 if ((ip->i_d.di_nlink == 0) && xfs_inode_is_filestream(ip))
1605 xfs_filestream_deassociate(ip);
1608 * If we previously truncated this file and removed old data
1609 * in the process, we want to initiate "early" writeout on
1610 * the last close. This is an attempt to combat the notorious
1611 * NULL files problem which is particularly noticeable from a
1612 * truncate down, buffered (re-)write (delalloc), followed by
1613 * a crash. What we are effectively doing here is
1614 * significantly reducing the time window where we'd otherwise
1615 * be exposed to that problem.
1617 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
1619 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
1620 if (VN_DIRTY(VFS_I(ip)) && ip->i_delayed_blks > 0) {
1621 error = -filemap_flush(VFS_I(ip)->i_mapping);
1628 if (ip->i_d.di_nlink == 0)
1631 if (xfs_can_free_eofblocks(ip, false)) {
1634 * If we can't get the iolock just skip truncating the blocks
1635 * past EOF because we could deadlock with the mmap_sem
1636 * otherwise. We'll get another chance to drop them once the
1637 * last reference to the inode is dropped, so we'll never leak
1638 * blocks permanently.
1640 * Further, check if the inode is being opened, written and
1641 * closed frequently and we have delayed allocation blocks
1642 * outstanding (e.g. streaming writes from the NFS server),
1643 * truncating the blocks past EOF will cause fragmentation to
1646 * In this case don't do the truncation, either, but we have to
1647 * be careful how we detect this case. Blocks beyond EOF show
1648 * up as i_delayed_blks even when the inode is clean, so we
1649 * need to truncate them away first before checking for a dirty
1650 * release. Hence on the first dirty close we will still remove
1651 * the speculative allocation, but after that we will leave it
1654 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
1657 error = xfs_free_eofblocks(mp, ip, true);
1658 if (error && error != EAGAIN)
1661 /* delalloc blocks after truncation means it really is dirty */
1662 if (ip->i_delayed_blks)
1663 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
1669 * xfs_inactive_truncate
1671 * Called to perform a truncate when an inode becomes unlinked.
1674 xfs_inactive_truncate(
1675 struct xfs_inode *ip)
1677 struct xfs_mount *mp = ip->i_mount;
1678 struct xfs_trans *tp;
1681 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1682 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
1684 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1685 xfs_trans_cancel(tp, 0);
1689 xfs_ilock(ip, XFS_ILOCK_EXCL);
1690 xfs_trans_ijoin(tp, ip, 0);
1693 * Log the inode size first to prevent stale data exposure in the event
1694 * of a system crash before the truncate completes. See the related
1695 * comment in xfs_setattr_size() for details.
1697 ip->i_d.di_size = 0;
1698 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1700 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
1702 goto error_trans_cancel;
1704 ASSERT(ip->i_d.di_nextents == 0);
1706 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1710 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1714 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1716 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1721 * xfs_inactive_ifree()
1723 * Perform the inode free when an inode is unlinked.
1727 struct xfs_inode *ip)
1729 xfs_bmap_free_t free_list;
1730 xfs_fsblock_t first_block;
1732 struct xfs_mount *mp = ip->i_mount;
1733 struct xfs_trans *tp;
1736 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1737 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_ifree, 0, 0);
1739 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1740 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES);
1744 xfs_ilock(ip, XFS_ILOCK_EXCL);
1745 xfs_trans_ijoin(tp, ip, 0);
1747 xfs_bmap_init(&free_list, &first_block);
1748 error = xfs_ifree(tp, ip, &free_list);
1751 * If we fail to free the inode, shut down. The cancel
1752 * might do that, we need to make sure. Otherwise the
1753 * inode might be lost for a long time or forever.
1755 if (!XFS_FORCED_SHUTDOWN(mp)) {
1756 xfs_notice(mp, "%s: xfs_ifree returned error %d",
1758 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1760 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1761 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1766 * Credit the quota account(s). The inode is gone.
1768 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1771 * Just ignore errors at this point. There is nothing we can
1772 * do except to try to keep going. Make sure it's not a silent
1775 error = xfs_bmap_finish(&tp, &free_list, &committed);
1777 xfs_notice(mp, "%s: xfs_bmap_finish returned error %d",
1779 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1781 xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
1784 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1791 * This is called when the vnode reference count for the vnode
1792 * goes to zero. If the file has been unlinked, then it must
1793 * now be truncated. Also, we clear all of the read-ahead state
1794 * kept for the inode here since the file is now closed.
1800 struct xfs_mount *mp;
1805 * If the inode is already free, then there can be nothing
1808 if (ip->i_d.di_mode == 0) {
1809 ASSERT(ip->i_df.if_real_bytes == 0);
1810 ASSERT(ip->i_df.if_broot_bytes == 0);
1816 /* If this is a read-only mount, don't do this (would generate I/O) */
1817 if (mp->m_flags & XFS_MOUNT_RDONLY)
1820 if (ip->i_d.di_nlink != 0) {
1822 * force is true because we are evicting an inode from the
1823 * cache. Post-eof blocks must be freed, lest we end up with
1824 * broken free space accounting.
1826 if (xfs_can_free_eofblocks(ip, true))
1827 xfs_free_eofblocks(mp, ip, false);
1832 if (S_ISREG(ip->i_d.di_mode) &&
1833 (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 ||
1834 ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0))
1837 error = xfs_qm_dqattach(ip, 0);
1841 if (S_ISLNK(ip->i_d.di_mode))
1842 error = xfs_inactive_symlink(ip);
1844 error = xfs_inactive_truncate(ip);
1849 * If there are attributes associated with the file then blow them away
1850 * now. The code calls a routine that recursively deconstructs the
1851 * attribute fork. We need to just commit the current transaction
1852 * because we can't use it for xfs_attr_inactive().
1854 if (ip->i_d.di_anextents > 0) {
1855 ASSERT(ip->i_d.di_forkoff != 0);
1857 error = xfs_attr_inactive(ip);
1863 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1865 ASSERT(ip->i_d.di_anextents == 0);
1870 error = xfs_inactive_ifree(ip);
1875 * Release the dquots held by inode, if any.
1877 xfs_qm_dqdetach(ip);
1881 * This is called when the inode's link count goes to 0.
1882 * We place the on-disk inode on a list in the AGI. It
1883 * will be pulled from this list when the inode is freed.
1900 ASSERT(ip->i_d.di_nlink == 0);
1901 ASSERT(ip->i_d.di_mode != 0);
1906 * Get the agi buffer first. It ensures lock ordering
1909 error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp);
1912 agi = XFS_BUF_TO_AGI(agibp);
1915 * Get the index into the agi hash table for the
1916 * list this inode will go on.
1918 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
1920 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
1921 ASSERT(agi->agi_unlinked[bucket_index]);
1922 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino);
1924 if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) {
1926 * There is already another inode in the bucket we need
1927 * to add ourselves to. Add us at the front of the list.
1928 * Here we put the head pointer into our next pointer,
1929 * and then we fall through to point the head at us.
1931 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
1936 ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO));
1937 dip->di_next_unlinked = agi->agi_unlinked[bucket_index];
1938 offset = ip->i_imap.im_boffset +
1939 offsetof(xfs_dinode_t, di_next_unlinked);
1941 /* need to recalc the inode CRC if appropriate */
1942 xfs_dinode_calc_crc(mp, dip);
1944 xfs_trans_inode_buf(tp, ibp);
1945 xfs_trans_log_buf(tp, ibp, offset,
1946 (offset + sizeof(xfs_agino_t) - 1));
1947 xfs_inobp_check(mp, ibp);
1951 * Point the bucket head pointer at the inode being inserted.
1954 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino);
1955 offset = offsetof(xfs_agi_t, agi_unlinked) +
1956 (sizeof(xfs_agino_t) * bucket_index);
1957 xfs_trans_log_buf(tp, agibp, offset,
1958 (offset + sizeof(xfs_agino_t) - 1));
1963 * Pull the on-disk inode from the AGI unlinked list.
1976 xfs_agnumber_t agno;
1978 xfs_agino_t next_agino;
1979 xfs_buf_t *last_ibp;
1980 xfs_dinode_t *last_dip = NULL;
1982 int offset, last_offset = 0;
1986 agno = XFS_INO_TO_AGNO(mp, ip->i_ino);
1989 * Get the agi buffer first. It ensures lock ordering
1992 error = xfs_read_agi(mp, tp, agno, &agibp);
1996 agi = XFS_BUF_TO_AGI(agibp);
1999 * Get the index into the agi hash table for the
2000 * list this inode will go on.
2002 agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
2004 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
2005 ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO));
2006 ASSERT(agi->agi_unlinked[bucket_index]);
2008 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) {
2010 * We're at the head of the list. Get the inode's on-disk
2011 * buffer to see if there is anyone after us on the list.
2012 * Only modify our next pointer if it is not already NULLAGINO.
2013 * This saves us the overhead of dealing with the buffer when
2014 * there is no need to change it.
2016 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2019 xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.",
2023 next_agino = be32_to_cpu(dip->di_next_unlinked);
2024 ASSERT(next_agino != 0);
2025 if (next_agino != NULLAGINO) {
2026 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2027 offset = ip->i_imap.im_boffset +
2028 offsetof(xfs_dinode_t, di_next_unlinked);
2030 /* need to recalc the inode CRC if appropriate */
2031 xfs_dinode_calc_crc(mp, dip);
2033 xfs_trans_inode_buf(tp, ibp);
2034 xfs_trans_log_buf(tp, ibp, offset,
2035 (offset + sizeof(xfs_agino_t) - 1));
2036 xfs_inobp_check(mp, ibp);
2038 xfs_trans_brelse(tp, ibp);
2041 * Point the bucket head pointer at the next inode.
2043 ASSERT(next_agino != 0);
2044 ASSERT(next_agino != agino);
2045 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino);
2046 offset = offsetof(xfs_agi_t, agi_unlinked) +
2047 (sizeof(xfs_agino_t) * bucket_index);
2048 xfs_trans_log_buf(tp, agibp, offset,
2049 (offset + sizeof(xfs_agino_t) - 1));
2052 * We need to search the list for the inode being freed.
2054 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
2056 while (next_agino != agino) {
2057 struct xfs_imap imap;
2060 xfs_trans_brelse(tp, last_ibp);
2063 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino);
2065 error = xfs_imap(mp, tp, next_ino, &imap, 0);
2068 "%s: xfs_imap returned error %d.",
2073 error = xfs_imap_to_bp(mp, tp, &imap, &last_dip,
2077 "%s: xfs_imap_to_bp returned error %d.",
2082 last_offset = imap.im_boffset;
2083 next_agino = be32_to_cpu(last_dip->di_next_unlinked);
2084 ASSERT(next_agino != NULLAGINO);
2085 ASSERT(next_agino != 0);
2089 * Now last_ibp points to the buffer previous to us on the
2090 * unlinked list. Pull us from the list.
2092 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp,
2095 xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.",
2099 next_agino = be32_to_cpu(dip->di_next_unlinked);
2100 ASSERT(next_agino != 0);
2101 ASSERT(next_agino != agino);
2102 if (next_agino != NULLAGINO) {
2103 dip->di_next_unlinked = cpu_to_be32(NULLAGINO);
2104 offset = ip->i_imap.im_boffset +
2105 offsetof(xfs_dinode_t, di_next_unlinked);
2107 /* need to recalc the inode CRC if appropriate */
2108 xfs_dinode_calc_crc(mp, dip);
2110 xfs_trans_inode_buf(tp, ibp);
2111 xfs_trans_log_buf(tp, ibp, offset,
2112 (offset + sizeof(xfs_agino_t) - 1));
2113 xfs_inobp_check(mp, ibp);
2115 xfs_trans_brelse(tp, ibp);
2118 * Point the previous inode on the list to the next inode.
2120 last_dip->di_next_unlinked = cpu_to_be32(next_agino);
2121 ASSERT(next_agino != 0);
2122 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked);
2124 /* need to recalc the inode CRC if appropriate */
2125 xfs_dinode_calc_crc(mp, last_dip);
2127 xfs_trans_inode_buf(tp, last_ibp);
2128 xfs_trans_log_buf(tp, last_ibp, offset,
2129 (offset + sizeof(xfs_agino_t) - 1));
2130 xfs_inobp_check(mp, last_ibp);
2136 * A big issue when freeing the inode cluster is that we _cannot_ skip any
2137 * inodes that are in memory - they all must be marked stale and attached to
2138 * the cluster buffer.
2142 xfs_inode_t *free_ip,
2146 xfs_mount_t *mp = free_ip->i_mount;
2147 int blks_per_cluster;
2154 xfs_inode_log_item_t *iip;
2155 xfs_log_item_t *lip;
2156 struct xfs_perag *pag;
2158 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum));
2159 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) {
2160 blks_per_cluster = 1;
2161 ninodes = mp->m_sb.sb_inopblock;
2162 nbufs = XFS_IALLOC_BLOCKS(mp);
2164 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) /
2165 mp->m_sb.sb_blocksize;
2166 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock;
2167 nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster;
2170 for (j = 0; j < nbufs; j++, inum += ninodes) {
2171 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
2172 XFS_INO_TO_AGBNO(mp, inum));
2175 * We obtain and lock the backing buffer first in the process
2176 * here, as we have to ensure that any dirty inode that we
2177 * can't get the flush lock on is attached to the buffer.
2178 * If we scan the in-memory inodes first, then buffer IO can
2179 * complete before we get a lock on it, and hence we may fail
2180 * to mark all the active inodes on the buffer stale.
2182 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
2183 mp->m_bsize * blks_per_cluster,
2190 * This buffer may not have been correctly initialised as we
2191 * didn't read it from disk. That's not important because we are
2192 * only using to mark the buffer as stale in the log, and to
2193 * attach stale cached inodes on it. That means it will never be
2194 * dispatched for IO. If it is, we want to know about it, and we
2195 * want it to fail. We can acheive this by adding a write
2196 * verifier to the buffer.
2198 bp->b_ops = &xfs_inode_buf_ops;
2201 * Walk the inodes already attached to the buffer and mark them
2202 * stale. These will all have the flush locks held, so an
2203 * in-memory inode walk can't lock them. By marking them all
2204 * stale first, we will not attempt to lock them in the loop
2205 * below as the XFS_ISTALE flag will be set.
2209 if (lip->li_type == XFS_LI_INODE) {
2210 iip = (xfs_inode_log_item_t *)lip;
2211 ASSERT(iip->ili_logged == 1);
2212 lip->li_cb = xfs_istale_done;
2213 xfs_trans_ail_copy_lsn(mp->m_ail,
2214 &iip->ili_flush_lsn,
2215 &iip->ili_item.li_lsn);
2216 xfs_iflags_set(iip->ili_inode, XFS_ISTALE);
2218 lip = lip->li_bio_list;
2223 * For each inode in memory attempt to add it to the inode
2224 * buffer and set it up for being staled on buffer IO
2225 * completion. This is safe as we've locked out tail pushing
2226 * and flushing by locking the buffer.
2228 * We have already marked every inode that was part of a
2229 * transaction stale above, which means there is no point in
2230 * even trying to lock them.
2232 for (i = 0; i < ninodes; i++) {
2235 ip = radix_tree_lookup(&pag->pag_ici_root,
2236 XFS_INO_TO_AGINO(mp, (inum + i)));
2238 /* Inode not in memory, nothing to do */
2245 * because this is an RCU protected lookup, we could
2246 * find a recently freed or even reallocated inode
2247 * during the lookup. We need to check under the
2248 * i_flags_lock for a valid inode here. Skip it if it
2249 * is not valid, the wrong inode or stale.
2251 spin_lock(&ip->i_flags_lock);
2252 if (ip->i_ino != inum + i ||
2253 __xfs_iflags_test(ip, XFS_ISTALE)) {
2254 spin_unlock(&ip->i_flags_lock);
2258 spin_unlock(&ip->i_flags_lock);
2261 * Don't try to lock/unlock the current inode, but we
2262 * _cannot_ skip the other inodes that we did not find
2263 * in the list attached to the buffer and are not
2264 * already marked stale. If we can't lock it, back off
2267 if (ip != free_ip &&
2268 !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2276 xfs_iflags_set(ip, XFS_ISTALE);
2279 * we don't need to attach clean inodes or those only
2280 * with unlogged changes (which we throw away, anyway).
2283 if (!iip || xfs_inode_clean(ip)) {
2284 ASSERT(ip != free_ip);
2286 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2290 iip->ili_last_fields = iip->ili_fields;
2291 iip->ili_fields = 0;
2292 iip->ili_logged = 1;
2293 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
2294 &iip->ili_item.li_lsn);
2296 xfs_buf_attach_iodone(bp, xfs_istale_done,
2300 xfs_iunlock(ip, XFS_ILOCK_EXCL);
2303 xfs_trans_stale_inode_buf(tp, bp);
2304 xfs_trans_binval(tp, bp);
2312 * This is called to return an inode to the inode free list.
2313 * The inode should already be truncated to 0 length and have
2314 * no pages associated with it. This routine also assumes that
2315 * the inode is already a part of the transaction.
2317 * The on-disk copy of the inode will have been added to the list
2318 * of unlinked inodes in the AGI. We need to remove the inode from
2319 * that list atomically with respect to freeing it here.
2325 xfs_bmap_free_t *flist)
2329 xfs_ino_t first_ino;
2331 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
2332 ASSERT(ip->i_d.di_nlink == 0);
2333 ASSERT(ip->i_d.di_nextents == 0);
2334 ASSERT(ip->i_d.di_anextents == 0);
2335 ASSERT(ip->i_d.di_size == 0 || !S_ISREG(ip->i_d.di_mode));
2336 ASSERT(ip->i_d.di_nblocks == 0);
2339 * Pull the on-disk inode from the AGI unlinked list.
2341 error = xfs_iunlink_remove(tp, ip);
2345 error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino);
2349 ip->i_d.di_mode = 0; /* mark incore inode as free */
2350 ip->i_d.di_flags = 0;
2351 ip->i_d.di_dmevmask = 0;
2352 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */
2353 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS;
2354 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
2356 * Bump the generation count so no one will be confused
2357 * by reincarnations of this inode.
2360 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
2363 error = xfs_ifree_cluster(ip, tp, first_ino);
2369 * This is called to unpin an inode. The caller must have the inode locked
2370 * in at least shared mode so that the buffer cannot be subsequently pinned
2371 * once someone is waiting for it to be unpinned.
2375 struct xfs_inode *ip)
2377 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
2379 trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
2381 /* Give the log a push to start the unpinning I/O */
2382 xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0);
2388 struct xfs_inode *ip)
2390 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
2391 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
2396 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2397 if (xfs_ipincount(ip))
2399 } while (xfs_ipincount(ip));
2400 finish_wait(wq, &wait.wait);
2405 struct xfs_inode *ip)
2407 if (xfs_ipincount(ip))
2408 __xfs_iunpin_wait(ip);
2414 struct xfs_name *name,
2417 xfs_mount_t *mp = dp->i_mount;
2418 xfs_trans_t *tp = NULL;
2419 int is_dir = S_ISDIR(ip->i_d.di_mode);
2421 xfs_bmap_free_t free_list;
2422 xfs_fsblock_t first_block;
2429 trace_xfs_remove(dp, name);
2431 if (XFS_FORCED_SHUTDOWN(mp))
2432 return XFS_ERROR(EIO);
2434 error = xfs_qm_dqattach(dp, 0);
2438 error = xfs_qm_dqattach(ip, 0);
2443 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
2444 log_count = XFS_DEFAULT_LOG_COUNT;
2446 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
2447 log_count = XFS_REMOVE_LOG_COUNT;
2449 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2452 * We try to get the real space reservation first,
2453 * allowing for directory btree deletion(s) implying
2454 * possible bmap insert(s). If we can't get the space
2455 * reservation then we use 0 instead, and avoid the bmap
2456 * btree insert(s) in the directory code by, if the bmap
2457 * insert tries to happen, instead trimming the LAST
2458 * block from the directory.
2460 resblks = XFS_REMOVE_SPACE_RES(mp);
2461 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_remove, resblks, 0);
2462 if (error == ENOSPC) {
2464 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_remove, 0, 0);
2467 ASSERT(error != ENOSPC);
2469 goto out_trans_cancel;
2472 xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL);
2474 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2475 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2478 * If we're removing a directory perform some additional validation.
2481 ASSERT(ip->i_d.di_nlink >= 2);
2482 if (ip->i_d.di_nlink != 2) {
2483 error = XFS_ERROR(ENOTEMPTY);
2484 goto out_trans_cancel;
2486 if (!xfs_dir_isempty(ip)) {
2487 error = XFS_ERROR(ENOTEMPTY);
2488 goto out_trans_cancel;
2492 xfs_bmap_init(&free_list, &first_block);
2493 error = xfs_dir_removename(tp, dp, name, ip->i_ino,
2494 &first_block, &free_list, resblks);
2496 ASSERT(error != ENOENT);
2497 goto out_bmap_cancel;
2499 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2503 * Drop the link from ip's "..".
2505 error = xfs_droplink(tp, dp);
2507 goto out_bmap_cancel;
2510 * Drop the "." link from ip to self.
2512 error = xfs_droplink(tp, ip);
2514 goto out_bmap_cancel;
2517 * When removing a non-directory we need to log the parent
2518 * inode here. For a directory this is done implicitly
2519 * by the xfs_droplink call for the ".." entry.
2521 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2525 * Drop the link from dp to ip.
2527 error = xfs_droplink(tp, ip);
2529 goto out_bmap_cancel;
2532 * Determine if this is the last link while
2533 * we are in the transaction.
2535 link_zero = (ip->i_d.di_nlink == 0);
2538 * If this is a synchronous mount, make sure that the
2539 * remove transaction goes to disk before returning to
2542 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
2543 xfs_trans_set_sync(tp);
2545 error = xfs_bmap_finish(&tp, &free_list, &committed);
2547 goto out_bmap_cancel;
2549 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2554 * If we are using filestreams, kill the stream association.
2555 * If the file is still open it may get a new one but that
2556 * will get killed on last close in xfs_close() so we don't
2557 * have to worry about that.
2559 if (!is_dir && link_zero && xfs_inode_is_filestream(ip))
2560 xfs_filestream_deassociate(ip);
2565 xfs_bmap_cancel(&free_list);
2566 cancel_flags |= XFS_TRANS_ABORT;
2568 xfs_trans_cancel(tp, cancel_flags);
2574 * Enter all inodes for a rename transaction into a sorted array.
2577 xfs_sort_for_rename(
2578 xfs_inode_t *dp1, /* in: old (source) directory inode */
2579 xfs_inode_t *dp2, /* in: new (target) directory inode */
2580 xfs_inode_t *ip1, /* in: inode of old entry */
2581 xfs_inode_t *ip2, /* in: inode of new entry, if it
2582 already exists, NULL otherwise. */
2583 xfs_inode_t **i_tab,/* out: array of inode returned, sorted */
2584 int *num_inodes) /* out: number of inodes in array */
2590 * i_tab contains a list of pointers to inodes. We initialize
2591 * the table here & we'll sort it. We will then use it to
2592 * order the acquisition of the inode locks.
2594 * Note that the table may contain duplicates. e.g., dp1 == dp2.
2608 * Sort the elements via bubble sort. (Remember, there are at
2609 * most 4 elements to sort, so this is adequate.)
2611 for (i = 0; i < *num_inodes; i++) {
2612 for (j = 1; j < *num_inodes; j++) {
2613 if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
2615 i_tab[j] = i_tab[j-1];
2627 xfs_inode_t *src_dp,
2628 struct xfs_name *src_name,
2629 xfs_inode_t *src_ip,
2630 xfs_inode_t *target_dp,
2631 struct xfs_name *target_name,
2632 xfs_inode_t *target_ip)
2634 xfs_trans_t *tp = NULL;
2635 xfs_mount_t *mp = src_dp->i_mount;
2636 int new_parent; /* moving to a new dir */
2637 int src_is_directory; /* src_name is a directory */
2639 xfs_bmap_free_t free_list;
2640 xfs_fsblock_t first_block;
2643 xfs_inode_t *inodes[4];
2647 trace_xfs_rename(src_dp, target_dp, src_name, target_name);
2649 new_parent = (src_dp != target_dp);
2650 src_is_directory = S_ISDIR(src_ip->i_d.di_mode);
2652 xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip,
2653 inodes, &num_inodes);
2655 xfs_bmap_init(&free_list, &first_block);
2656 tp = xfs_trans_alloc(mp, XFS_TRANS_RENAME);
2657 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2658 spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len);
2659 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_rename, spaceres, 0);
2660 if (error == ENOSPC) {
2662 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_rename, 0, 0);
2665 xfs_trans_cancel(tp, 0);
2670 * Attach the dquots to the inodes
2672 error = xfs_qm_vop_rename_dqattach(inodes);
2674 xfs_trans_cancel(tp, cancel_flags);
2679 * Lock all the participating inodes. Depending upon whether
2680 * the target_name exists in the target directory, and
2681 * whether the target directory is the same as the source
2682 * directory, we can lock from 2 to 4 inodes.
2684 xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
2687 * Join all the inodes to the transaction. From this point on,
2688 * we can rely on either trans_commit or trans_cancel to unlock
2691 xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL);
2693 xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL);
2694 xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
2696 xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
2699 * If we are using project inheritance, we only allow renames
2700 * into our tree when the project IDs are the same; else the
2701 * tree quota mechanism would be circumvented.
2703 if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
2704 (xfs_get_projid(target_dp) != xfs_get_projid(src_ip)))) {
2705 error = XFS_ERROR(EXDEV);
2710 * Set up the target.
2712 if (target_ip == NULL) {
2714 * If there's no space reservation, check the entry will
2715 * fit before actually inserting it.
2717 error = xfs_dir_canenter(tp, target_dp, target_name, spaceres);
2721 * If target does not exist and the rename crosses
2722 * directories, adjust the target directory link count
2723 * to account for the ".." reference from the new entry.
2725 error = xfs_dir_createname(tp, target_dp, target_name,
2726 src_ip->i_ino, &first_block,
2727 &free_list, spaceres);
2728 if (error == ENOSPC)
2733 xfs_trans_ichgtime(tp, target_dp,
2734 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2736 if (new_parent && src_is_directory) {
2737 error = xfs_bumplink(tp, target_dp);
2741 } else { /* target_ip != NULL */
2743 * If target exists and it's a directory, check that both
2744 * target and source are directories and that target can be
2745 * destroyed, or that neither is a directory.
2747 if (S_ISDIR(target_ip->i_d.di_mode)) {
2749 * Make sure target dir is empty.
2751 if (!(xfs_dir_isempty(target_ip)) ||
2752 (target_ip->i_d.di_nlink > 2)) {
2753 error = XFS_ERROR(EEXIST);
2759 * Link the source inode under the target name.
2760 * If the source inode is a directory and we are moving
2761 * it across directories, its ".." entry will be
2762 * inconsistent until we replace that down below.
2764 * In case there is already an entry with the same
2765 * name at the destination directory, remove it first.
2767 error = xfs_dir_replace(tp, target_dp, target_name,
2769 &first_block, &free_list, spaceres);
2773 xfs_trans_ichgtime(tp, target_dp,
2774 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2777 * Decrement the link count on the target since the target
2778 * dir no longer points to it.
2780 error = xfs_droplink(tp, target_ip);
2784 if (src_is_directory) {
2786 * Drop the link from the old "." entry.
2788 error = xfs_droplink(tp, target_ip);
2792 } /* target_ip != NULL */
2795 * Remove the source.
2797 if (new_parent && src_is_directory) {
2799 * Rewrite the ".." entry to point to the new
2802 error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot,
2804 &first_block, &free_list, spaceres);
2805 ASSERT(error != EEXIST);
2811 * We always want to hit the ctime on the source inode.
2813 * This isn't strictly required by the standards since the source
2814 * inode isn't really being changed, but old unix file systems did
2815 * it and some incremental backup programs won't work without it.
2817 xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG);
2818 xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE);
2821 * Adjust the link count on src_dp. This is necessary when
2822 * renaming a directory, either within one parent when
2823 * the target existed, or across two parent directories.
2825 if (src_is_directory && (new_parent || target_ip != NULL)) {
2828 * Decrement link count on src_directory since the
2829 * entry that's moved no longer points to it.
2831 error = xfs_droplink(tp, src_dp);
2836 error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino,
2837 &first_block, &free_list, spaceres);
2841 xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2842 xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
2844 xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
2847 * If this is a synchronous mount, make sure that the
2848 * rename transaction goes to disk before returning to
2851 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2852 xfs_trans_set_sync(tp);
2855 error = xfs_bmap_finish(&tp, &free_list, &committed);
2857 xfs_bmap_cancel(&free_list);
2858 xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES |
2864 * trans_commit will unlock src_ip, target_ip & decrement
2865 * the vnode references.
2867 return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
2870 cancel_flags |= XFS_TRANS_ABORT;
2872 xfs_bmap_cancel(&free_list);
2873 xfs_trans_cancel(tp, cancel_flags);
2883 xfs_mount_t *mp = ip->i_mount;
2884 struct xfs_perag *pag;
2885 unsigned long first_index, mask;
2886 unsigned long inodes_per_cluster;
2888 xfs_inode_t **ilist;
2895 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
2897 inodes_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog;
2898 ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *);
2899 ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS);
2903 mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1);
2904 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask;
2906 /* really need a gang lookup range call here */
2907 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist,
2908 first_index, inodes_per_cluster);
2912 for (i = 0; i < nr_found; i++) {
2918 * because this is an RCU protected lookup, we could find a
2919 * recently freed or even reallocated inode during the lookup.
2920 * We need to check under the i_flags_lock for a valid inode
2921 * here. Skip it if it is not valid or the wrong inode.
2923 spin_lock(&ip->i_flags_lock);
2925 (XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index) {
2926 spin_unlock(&ip->i_flags_lock);
2929 spin_unlock(&ip->i_flags_lock);
2932 * Do an un-protected check to see if the inode is dirty and
2933 * is a candidate for flushing. These checks will be repeated
2934 * later after the appropriate locks are acquired.
2936 if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0)
2940 * Try to get locks. If any are unavailable or it is pinned,
2941 * then this inode cannot be flushed and is skipped.
2944 if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED))
2946 if (!xfs_iflock_nowait(iq)) {
2947 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2950 if (xfs_ipincount(iq)) {
2952 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2957 * arriving here means that this inode can be flushed. First
2958 * re-check that it's dirty before flushing.
2960 if (!xfs_inode_clean(iq)) {
2962 error = xfs_iflush_int(iq, bp);
2964 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2965 goto cluster_corrupt_out;
2971 xfs_iunlock(iq, XFS_ILOCK_SHARED);
2975 XFS_STATS_INC(xs_icluster_flushcnt);
2976 XFS_STATS_ADD(xs_icluster_flushinode, clcount);
2987 cluster_corrupt_out:
2989 * Corruption detected in the clustering loop. Invalidate the
2990 * inode buffer and shut down the filesystem.
2994 * Clean up the buffer. If it was delwri, just release it --
2995 * brelse can handle it with no problems. If not, shut down the
2996 * filesystem before releasing the buffer.
2998 bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q);
3002 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3004 if (!bufwasdelwri) {
3006 * Just like incore_relse: if we have b_iodone functions,
3007 * mark the buffer as an error and call them. Otherwise
3008 * mark it as stale and brelse.
3013 xfs_buf_ioerror(bp, EIO);
3014 xfs_buf_ioend(bp, 0);
3022 * Unlocks the flush lock
3024 xfs_iflush_abort(iq, false);
3027 return XFS_ERROR(EFSCORRUPTED);
3031 * Flush dirty inode metadata into the backing buffer.
3033 * The caller must have the inode lock and the inode flush lock held. The
3034 * inode lock will still be held upon return to the caller, and the inode
3035 * flush lock will be released after the inode has reached the disk.
3037 * The caller must write out the buffer returned in *bpp and release it.
3041 struct xfs_inode *ip,
3042 struct xfs_buf **bpp)
3044 struct xfs_mount *mp = ip->i_mount;
3046 struct xfs_dinode *dip;
3049 XFS_STATS_INC(xs_iflush_count);
3051 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3052 ASSERT(xfs_isiflocked(ip));
3053 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3054 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3058 xfs_iunpin_wait(ip);
3061 * For stale inodes we cannot rely on the backing buffer remaining
3062 * stale in cache for the remaining life of the stale inode and so
3063 * xfs_imap_to_bp() below may give us a buffer that no longer contains
3064 * inodes below. We have to check this after ensuring the inode is
3065 * unpinned so that it is safe to reclaim the stale inode after the
3068 if (xfs_iflags_test(ip, XFS_ISTALE)) {
3074 * This may have been unpinned because the filesystem is shutting
3075 * down forcibly. If that's the case we must not write this inode
3076 * to disk, because the log record didn't make it to disk.
3078 * We also have to remove the log item from the AIL in this case,
3079 * as we wait for an empty AIL as part of the unmount process.
3081 if (XFS_FORCED_SHUTDOWN(mp)) {
3082 error = XFS_ERROR(EIO);
3087 * Get the buffer containing the on-disk inode.
3089 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK,
3097 * First flush out the inode that xfs_iflush was called with.
3099 error = xfs_iflush_int(ip, bp);
3104 * If the buffer is pinned then push on the log now so we won't
3105 * get stuck waiting in the write for too long.
3107 if (xfs_buf_ispinned(bp))
3108 xfs_log_force(mp, 0);
3112 * see if other inodes can be gathered into this write
3114 error = xfs_iflush_cluster(ip, bp);
3116 goto cluster_corrupt_out;
3123 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
3124 cluster_corrupt_out:
3125 error = XFS_ERROR(EFSCORRUPTED);
3128 * Unlocks the flush lock
3130 xfs_iflush_abort(ip, false);
3136 struct xfs_inode *ip,
3139 struct xfs_inode_log_item *iip = ip->i_itemp;
3140 struct xfs_dinode *dip;
3141 struct xfs_mount *mp = ip->i_mount;
3143 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
3144 ASSERT(xfs_isiflocked(ip));
3145 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
3146 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
3147 ASSERT(iip != NULL && iip->ili_fields != 0);
3149 /* set *dip = inode's place in the buffer */
3150 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset);
3152 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
3153 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) {
3154 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3155 "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p",
3156 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
3159 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC,
3160 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) {
3161 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3162 "%s: Bad inode %Lu, ptr 0x%p, magic number 0x%x",
3163 __func__, ip->i_ino, ip, ip->i_d.di_magic);
3166 if (S_ISREG(ip->i_d.di_mode)) {
3168 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3169 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE),
3170 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) {
3171 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3172 "%s: Bad regular inode %Lu, ptr 0x%p",
3173 __func__, ip->i_ino, ip);
3176 } else if (S_ISDIR(ip->i_d.di_mode)) {
3178 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) &&
3179 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
3180 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL),
3181 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) {
3182 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3183 "%s: Bad directory inode %Lu, ptr 0x%p",
3184 __func__, ip->i_ino, ip);
3188 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents >
3189 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5,
3190 XFS_RANDOM_IFLUSH_5)) {
3191 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3192 "%s: detected corrupt incore inode %Lu, "
3193 "total extents = %d, nblocks = %Ld, ptr 0x%p",
3194 __func__, ip->i_ino,
3195 ip->i_d.di_nextents + ip->i_d.di_anextents,
3196 ip->i_d.di_nblocks, ip);
3199 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize,
3200 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) {
3201 xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
3202 "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p",
3203 __func__, ip->i_ino, ip->i_d.di_forkoff, ip);
3208 * Inode item log recovery for v1/v2 inodes are dependent on the
3209 * di_flushiter count for correct sequencing. We bump the flush
3210 * iteration count so we can detect flushes which postdate a log record
3211 * during recovery. This is redundant as we now log every change and
3212 * hence this can't happen but we need to still do it to ensure
3213 * backwards compatibility with old kernels that predate logging all
3216 if (ip->i_d.di_version < 3)
3217 ip->i_d.di_flushiter++;
3220 * Copy the dirty parts of the inode into the on-disk
3221 * inode. We always copy out the core of the inode,
3222 * because if the inode is dirty at all the core must
3225 xfs_dinode_to_disk(dip, &ip->i_d);
3227 /* Wrap, we never let the log put out DI_MAX_FLUSH */
3228 if (ip->i_d.di_flushiter == DI_MAX_FLUSH)
3229 ip->i_d.di_flushiter = 0;
3232 * If this is really an old format inode and the superblock version
3233 * has not been updated to support only new format inodes, then
3234 * convert back to the old inode format. If the superblock version
3235 * has been updated, then make the conversion permanent.
3237 ASSERT(ip->i_d.di_version == 1 || xfs_sb_version_hasnlink(&mp->m_sb));
3238 if (ip->i_d.di_version == 1) {
3239 if (!xfs_sb_version_hasnlink(&mp->m_sb)) {
3243 ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1);
3244 dip->di_onlink = cpu_to_be16(ip->i_d.di_nlink);
3247 * The superblock version has already been bumped,
3248 * so just make the conversion to the new inode
3251 ip->i_d.di_version = 2;
3252 dip->di_version = 2;
3253 ip->i_d.di_onlink = 0;
3255 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad));
3256 memset(&(dip->di_pad[0]), 0,
3257 sizeof(dip->di_pad));
3258 ASSERT(xfs_get_projid(ip) == 0);
3262 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp);
3263 if (XFS_IFORK_Q(ip))
3264 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp);
3265 xfs_inobp_check(mp, bp);
3268 * We've recorded everything logged in the inode, so we'd like to clear
3269 * the ili_fields bits so we don't log and flush things unnecessarily.
3270 * However, we can't stop logging all this information until the data
3271 * we've copied into the disk buffer is written to disk. If we did we
3272 * might overwrite the copy of the inode in the log with all the data
3273 * after re-logging only part of it, and in the face of a crash we
3274 * wouldn't have all the data we need to recover.
3276 * What we do is move the bits to the ili_last_fields field. When
3277 * logging the inode, these bits are moved back to the ili_fields field.
3278 * In the xfs_iflush_done() routine we clear ili_last_fields, since we
3279 * know that the information those bits represent is permanently on
3280 * disk. As long as the flush completes before the inode is logged
3281 * again, then both ili_fields and ili_last_fields will be cleared.
3283 * We can play with the ili_fields bits here, because the inode lock
3284 * must be held exclusively in order to set bits there and the flush
3285 * lock protects the ili_last_fields bits. Set ili_logged so the flush
3286 * done routine can tell whether or not to look in the AIL. Also, store
3287 * the current LSN of the inode so that we can tell whether the item has
3288 * moved in the AIL from xfs_iflush_done(). In order to read the lsn we
3289 * need the AIL lock, because it is a 64 bit value that cannot be read
3292 iip->ili_last_fields = iip->ili_fields;
3293 iip->ili_fields = 0;
3294 iip->ili_logged = 1;
3296 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
3297 &iip->ili_item.li_lsn);
3300 * Attach the function xfs_iflush_done to the inode's
3301 * buffer. This will remove the inode from the AIL
3302 * and unlock the inode's flush lock when the inode is
3303 * completely written to disk.
3305 xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item);
3307 /* update the lsn in the on disk inode if required */
3308 if (ip->i_d.di_version == 3)
3309 dip->di_lsn = cpu_to_be64(iip->ili_item.li_lsn);
3311 /* generate the checksum. */
3312 xfs_dinode_calc_crc(mp, dip);
3314 ASSERT(bp->b_fspriv != NULL);
3315 ASSERT(bp->b_iodone != NULL);
3319 return XFS_ERROR(EFSCORRUPTED);